quimmedes/Deepwen-3.6

Deepwen 3.6

Deepwen 3.6 is a fine-tuned derivative of Qwen/Qwen3.6-35B-A3B (Mixture-of-Experts, ~35B total / ~3B active), specialized for AAA GameDev 3D production workflows: procedural geometry, hard-surface shape language, and Blender/Unreal asset pipelines.

What the model has

• Advanced thinking (DeepSeek style) — the model reasons before it answers. Its thinking comes from two sources:
  • Supervised reasoning training: 93.6% of its reasoning-focused training examples (103/110) carry a full reasoning chain as part of the target.
  • Reasoning-effort control: a chat template ported from deepseek-ai/DeepSeek-V4-Flash-0731, with three effort levels — low (default), xhigh, and max ("Beyond maximum — exhaustive, relentless... do not stop reasoning until you have independently verified the solution from multiple angles").
• Procedural 3D generation — explicit blockout gating before high-poly, conditional lightmap workflows, combinatorial validation, non-destructive pipelines.
• Hard-surface shape language — stance/relational design, primary volume architecture, motif propagation, panel breakup.
• Multi-skill asset workflows — Blender modifier-driven gear recipes, tooth profile generation, PBR game-prep, layered lighting legibility.

Improvements over the base model

Paired evaluations on held-out tasks (same server, same seeds):

Capability	Improvement
Procedural generation	blockout_gate: PARTIAL → PASS; conditional_lightmap: FAIL → PASS; 5 units improved vs 2 regressed
Replay safety	base competence suite 6/6 intact (no regressions)
Shape / hard-surface	no catastrophic flips across held-out objects
Blender gear recipe	modifier_workflow, tooth_profile_generation, game_prep_uv_pbr, non_destructive_order
Lighting	layered lighting legibility (bounce and ambient)

Quantizations (MoQ)

All files quantized with the Mixture of Quantizations (MoQ) method proposed by Waleed Ahmad: per-tensor type selection (attention/embeddings at higher precision, MLP/experts at more aggressive types) instead of a single type for every tensor.

File	Approx. size	Notes
Deepwen-3.6-Q2-MoQ.gguf	~10 GB	aggressive MoQ mix
Deepwen-3.6-Q3-MoQ.gguf	~13 GB	3.0 bpw target
Deepwen-3.6-Q4.5-MoQ.gguf	~21 GB	4.5 bpw target, sweet spot for local use
Deepwen-3.6-Q5-MoQ.gguf	~22 GB	5.0 bpw target
Deepwen-3.6-Q6-MoQ.gguf	~27 GB
Deepwen-3.6-Q8-MoQ.gguf	~36 GB	near-lossless
Deepwen-3.6-BF16.gguf	~70 GB	original merged weights, bf16

Usage

llama-server -m Deepwen-3.6-Q4.5-MoQ.gguf --host 0.0.0.0 --port 8080
# OpenAI-compatible: /v1/chat/completions

To enable advanced thinking at maximum effort:

{
  "messages": [{"role": "user", "content": "..."}],
  "chat_template_kwargs": {"reasoning_effort": "max"}
}

Works with llama.cpp (b3050+), LM Studio, Ollama, Jan.

Disclosures

• Base model: Qwen/Qwen3.6-35B-A3B — Copyright © Alibaba Group / Qwen Team. All rights to the base model and its weights remain with the original authors.
• The base model is released under the Apache License 2.0; this derivative inherits that license.
• Qwen 3.6 is a copyrighted, trademarked model family of Alibaba. "Deepwen 3.6" is an independent fine-tuned derivative and is not affiliated with, endorsed by, or sponsored by Alibaba / Qwen. The "Qwen" name is used solely to identify the base model.
• The embedded reasoning-effort prompts are adapted from the chat template of deepseek-ai/DeepSeek-V4-Flash-0731; DeepSeek remains the copyright holder of those prompt texts.
• MoQ quantization method: "Mixture of Quantizations" proposed by Waleed Ahmad (https://huggingface.co/w-ahmad).
• This model is provided as-is, without warranties of any kind, for research and local experimentation.